摘要
结冰探测在防除冰系统运行中起着至关重要的作用。本文提出了利用红外热波检测技术进行了积冰探测,并运用相关分析技术探讨了积冰边缘、厚度识别与冰形重建的方法。搭建了闪光脉冲红外主动式红外积冰探测实验平台,制备了规则型与阶跃型积冰样件,借助红外热像仪采集了受脉冲红外热激励后的积冰红外热信号。运用传统边缘检测方法与新构建的高斯‑拉普拉斯金字塔和面积滤波相结合的边缘检测算法进行了积冰边缘识别效果的对比与分析。利用积冰热信号的时空相关性,提出了在长短时记忆(Long short term memory,LSTM)模型中引入注意力机制建立端到端的红外探测积冰厚度预测模型(Convolutional neural netwok‑long short term memory‑efficient channel attention,CNN‑LSTM‑ECA),用以预测积冰厚度。此外,通过结合边缘检测和厚度预测,进行了阶梯状积冰样件的三维重建。结果表明,基于高斯‑拉普拉斯金字塔和区域滤波的传统边缘检测算法和新的边缘检测算法都可以用于检测冰的外边缘,但新算法在检测具有内部阶梯边界的冰边缘方面显示出显著的优势。基于信号特征的CNN‑LSTM‑ECA厚度预测模型在预测精度、稳定性和抗噪声性方面表现良好。重建三维积冰形状的数据来源于采集的数字信号和热图像,不受温度读数和传热条件的限制,具有更广阔的应用前景。此项研究为探索一种利用闪光脉冲红外技术进行积冰冰形有效、准确、定量识别提供可参考的方案。
Ice detection plays a crucial role in operation of anti/de-icing systems.An ice detection method exploiting infrared thermal wave detection technology was proposed,followed by the identification of the edge,thickness and shape reconstruction of ice accretion using the related processing techniques.An active infrared ice detection experimental platform was constructed by flash pulse infrared technology.With regular and step-shaped ice samples prepared,the infrared thermal signals of the ice accretion were captured by an infrared thermal imager.A comparative analysis of the edge detection effects was conducted between the traditional edge detection methods and a new edge detection algorithm combining Gaussian-Laplacian pyramids and area filtering.Through the spatiotemporal correlation of the ice thermal signal,an end-to-end infrared detection ice thickness prediction model(i.e.,convolutional neural network-long short term memory-efficient channel attention(CNN-LSTM-ECA))was constructed by introducing the attention mechanism into the LSTM model,with the thickness of the ice predicted.Further,three-dimensional reconstruction of ice accretion was performed by combining the edge detection and thickness prediction.It concluded that both traditional and new edge detection algorithms based on Gaussian-Laplacian pyramids and area filtering can be used to detect the outer edge of the ice,but the new algorithm shows a significant advantage in detecting the ice edge with internal step boundaries.The CNN-LSTM-ECA thickness prediction model based on signal features performs well in prediction accuracy,stability,and noise resistance.The data for reconstructing three-dimensional ice accretion shape come from the collected digital signals and thermal images,which are not limited by temperature reading and heat transfer conditions,and have a wider application prospect.This paper provides a reference for exploring an effective accurate and quantitative identification method for ice accretion detection based on flash pulse infrared thermography.
作者
李清英
勾一
刘森云
么娆
LI Qingying;GOU Yi;LIU Senyun;YAO Rao(School of Air Transportation,Shanghai University of Engineering Science,Shanghai 201620,P.R.China;Key Laboratory of Icing and Anti/De‑icing,China Aerodynamics Research and Development Center,Mianyang 621000,P.R.China)
基金
supported by the Open Fund of Key Laboratory of Icing and Anti/De-icing(No.IADL20200201)
the National Natural Science Founda-tion of China(No.62171271)。
关键词
积冰
红外探测
边缘检测
厚度预测
三维重建
ice accretion
infrared detection
edge detection
thickness prediction
three-dimensional shape reconstruction
